[ 
https://issues.apache.org/jira/browse/HDFS-13451?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16442046#comment-16442046
 ] 

lujie commented on HDFS-13451:
------------------------------

Hi  [~daryn]:

I strongly agree that removing the blindly is bad for postmodern debug.

I attach the new patch in which i try to replace  NPE with other exception 
which can give the real reason. For UGI  and  offline imag related NPE, I do 
not find suitable exception and they remains the same. And could you please 
review the new change in the new patch.

 

> Fix Some Potential NPE
> ----------------------
>
>                 Key: HDFS-13451
>                 URL: https://issues.apache.org/jira/browse/HDFS-13451
>             Project: Hadoop HDFS
>          Issue Type: Bug
>    Affects Versions: 3.0.0-beta1
>            Reporter: lujie
>            Priority: Major
>         Attachments: HDFS-13451_1.patch, HDFS-13451_2.patch
>
>
> We have developed a static analysis tool 
> [NPEDetector|https://github.com/lujiefsi/NPEDetector] to find some potential 
> NPE. Our analysis shows that some callees may return null in corner case(e.g. 
> node crash , IO exception), some of their callers have  _!=null_ check but 
> some do not have. In this issue we post a patch which can add  !=null  based 
> on existed !=null  check. For example:
> callee BlockInfo#getDatanode may return null:
> {code:java}
> public DatanodeDescriptor getDatanode(int index) {
>     DatanodeStorageInfo storage = getStorageInfo(index);
>    return storage == null ? null : storage.getDatanodeDescriptor();
> }
> {code}
> it has 4 callers, 3 of them have !=null checker, like in 
> CacheReplicationMonitor#addNewPendingCached:
> {code:java}
> DatanodeDescriptor datanode = blockInfo.getDatanode(i);
> if (datanode == null) {
>    continue;
> }
> {code}
> but in caller NamenodeFsck#blockIdCK have no !null checker, we add checker 
> just like CacheReplicationMonitor#addNewPendingCached
> {code:java}
> DatanodeDescriptor dn = blockInfo.getDatanode(idx);
> if (dn == null) {
>     continue;
> }
> {code}
> But due to we are not very  familiar with HDFS, hope some expert can review 
> it.
> Thanks!!!!



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to